Effects of Education on Fertility and...

THE WORLD BANK

Discussion Paper

EDUCATION AND TRAINING SERIES

Report No. EDT26

IThe Effects of Educationon Fertility and Mortality

Susan H. Cochrane

May,' 1986

Education and Training Department Operations Policy Staff

The views presented here are those of the author(s), and they should not be interpreted as reflecting those of the World Bank.

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Discussion Paper

Education and Training Series

Report No. EDT26

THE EFFECTS OF EDTJCATION ON FERTILITY AND MORTALITY

Susan H. Cochrane

Education Policy DivisionEducation and Training Department

May 1986

This is a two-part paper which contains "The Effects of Education onMortality: A Quick Review of the Evidence" and "The Effects of Educationand Urbanization on Fertility (Copyright 1983 Academic Press, reprintedw,ith permission)." The World Bank does not accept responsibility for theviews expressed herein, which are those of the author(s) and should not beattributed to the World Bank or to its affiliated organizations. Thefindings, interpretations, and conclusions are the results of research oranalysis supported by the Bank; they do not necessarily represent officialpolicy of the Bank.

(S 1986 The International Banlk for Reconstruction andDevelopment/The World Bank

Abstract

This volume consists of two reports, "The Effects of Education onMortality: A Quick Review of the Evidence," and "The Effects of Educationand Urbanization on Fertility."

The effects of education on mortality and fertility provide insightinto both the determinants of demographic phenomena and the socialconsequences of education. The evidence is clearest with respect to infantand child mortality. The higher the level of the parents' education, thelower the mortality of their children. The effect of education onfertility is somewhat more complicated. While fertility decreasesuniformly with education in some environments, there appears to be athreshold level of education and only at levels beyond primary school doesfertility decrease with increases in education.

The evidence presented in this paper tends to confirm earlier findingsthat parental education is inversely related to child mortality. Thisrelationship is somewhat stronger for women's education than men'seducation, as had been previously found. Evidence also confirms thateducation does not have a uniformly inverse relationship with fertility inall circumstances. In some environments, there is a threshold level ofeducation which must be achieved before fertility declines. Also, theeducation of women is more likely to be inversely related to fertility thanis the education of men. And finally, the more urbanized a country, themore likely education is to be uniformly inversely related to fertility.

PageNo.

THE EFFECTS OF EDUCATION ON MORTALITY:A QUICK REVIEW OF THE EVIDENCE 1

New Data ..... ...... 2Other Studies ...... ..... 3Conclusions........ ......................... 6Footnotes. .................. D*.*.s....e..c............. ^ 8

Bibliography ........... .*. *c.. e.** 9Table . .... e e 11

THE EFFECTS OF EDUCATION AND URBANIZATION ON FERTILITY 13

Measuring the Effects of Education and Residenceon Fertility: Methodological Issues ... . ...................... 16Measuring Eduation' s .... ............ eec...... 16Measuring Residence's Effect o e... ........ .ece 19Interactions, Nonlinearities and Measurement Problems......... 19

Literature Review ........ .... ...... ..... ccc...... 21Comparative Analysis of WFS Data.............e.. 22

Education ................... * ~~22R.esidence......................... ..cc..................... 26Interaction of Residence and Education9.ec.e.ee.eec.ce..ce.. 28

Other Studies of Education's Relationship to Fertility .....5. 29Meta ....................................................... 29Multivariate Studies Within Countriesc... .......... oe*006*09 30

Other Considerations............ ................ cc...... .... 33Stability of Educational Differentials Over Times..e..ec..ce... 33Other Evidence on Urban-Rural Differences in Fertility.ee cc... 35

Propositions e..c...ee.c.....cc...... ccc.... c.... cc.. 36Education and Fertility ............ ............ 36Residence and Fertility ........................ 37

Notes cc...... ... .... ................. ccc.. e cc ............ cc.... 38Bibiographye ........................ 4439Appendix Tal ............44

THE EFFECTS OF EDUCATION ON MORTALITY:A QUICK REVIEW OF THE EVIDENCE

THE EFFECTS OF EDUCATION ON FERTILITY AND MORTALITY

A Quick Review of the Evidence

In June of 1979 WHO sponsored a conference in Mexico City entitled"The Socioeconomic Determinants and Consequences of Mortality." A number ofpapers at that conference examined the relationship between national levels ofmortality and levels of development including measures of education. Inaddition, papers by Hugo Behm, John Caldwell and Paul Schultz analyzed theeffect of the education of parents on the mortality of their children. Allfound a strong effect of education on mortality in developing countries. Datafrom Latin America and Asia assembled elsewhere by Arriaga had also confirmeda strong inverse relationships between infant and child mortality and mother'seducation. (Arriaga 1979a and 1979b).

In late 1979 and early 1980 the World Bank sponsored a review ofboth the aggregate and the household level data on the relationship betweeneducation and mortality (Cochrane, O'Hara and Leslie, 1980). They used theexisting tabulations that could be found from censuses and surveys as well asreviewing multivariate analysis of household data sets. The findings werefairly striking. In the simple bivariate comparisons of parental educationand child survival they found virtually no deviation from the expectedpattern. The more schooling the mother or father had the lower the mortalityof their offspring. On average an additional year of mother's schooling wasassociated with a 9 per thousand reduction in child mortality.

The Bank researchers attempted to determine to the extent possiblewith existing studies, whether or not these results were spurious. First,they tested whether the effects were explained by urban, rural differences.Controlling for residence did not eliminate the effects of education. Secondthey attempted to determine if the relationship between wife's education andher child's mortality was the result of the fact that more educated women werein wealthier households married to more educated men. Crude tests indicatedthat only part of the effect of a mother's education could be the result ofthe fact that more educated women were married to more educated men. -Theeffects of husband's education were generally found to be about half as greatas the effect of mother's education. A review of the few multivariate studiesthat existed at that time led to the conclusion that of the 9 per thousandeffect, 6 per 1,000 was the result of the woman's own education and 3 per1,000 was the effect of the fact that she was married to a more educated,higher income husband.

-2-

New Data

Since the 1980 World Bank review, a considerable amount of work hasbeen done on the relationship between parental education and childmortality. One reason that it has been possible to do so much more work hasbeen because the WFS collected data on infant and child mortality on a widerange of developing countries. Table 1 shows comparisons of child survival byparental education characteristics of the family from the WFS data. This is avaluable addition to the earlier information on mortality and educationbecause it provides comparable data from 28 developing countries.

These data show a fairly uniform relationship between paternaleducation and child survival. There are a few deviation from the expectedpattern. If only sample sizes over 500 are examined there is only 1 deviationfrom the expected pattern for mother's education which exceeds 2% Forfather's education there are 6 deviations of that size or larger.- The datashow that on average under 5 mortality falls from 152 to 67 per 1,000 over therange of mothers education and from 148 to 84 per 1,000 over-the range ofhusband's education. Assuming the average education of the 7+ group is 9years that gives us a reduction of under 5 mortality of 9.4 per 1 000 per yearof schooling for mother's education and 7 per 1,000 for husband.2

These findings strongly confirm the results found in the earlierBank review: parental education has a close relationship with child survival,the strength of the relationship is greater for mother's education thanfather's education and the magnitude of the relationship is a reduction of 9per 1,000 for an additional year of mother's schooling.

The average relationship between education and child survival givesus some information. Deviations from that pattern are also informative. Inthe earlier study it was found that countries that had higher levels ofexpenditure per capita on public health tended to show smaller effects ofeducation on mortal ty, i.e., public health effects appeared to substitute foreducation effects.2

The regularity and strength of the relationship between parentaleducation and child health raise important questions of causation. Thesefindings could result from the fact that parental education is associated withother variables. This possibility was discussed briefly above. In additionif the association is not spurious, then it becomes necessary to understandthe channels through which education has its effects. Hobcraft, McDonald andRutstein (1984b) whose data were cited above have done analysis whichaddresses both the question of spurious association and in another paper theyprovide indirect insight into the question of channels.

In their paper on socio-economic differences in child survival, theyexamined five socio-economic variables; mother's education, her husband'seducation, current residence, her work status since marriage and her husband'soccupation. They estimated the best fit models with these 5 variables. Theirresults varied depending on the type of mortality analyzed. For neo-natal andpost neo-natal mortality, mother's education was the most important variablein a third to a fourth of the countries while it was signficant, if not themost important variable, in another third to a half of the cases. Husband's

education was important in approximately the same proportion of the cases.They found that the effect of mother's and husband's education gained inimportance as the child aged. For mortality from ages 1 to 5, they foundmother's education was included in the best model in 19 out of 20 cases andhusband's education in 16 of 26 cases. Thus, they confirmed the importance ofparental education controlling for other socioeconomic variables in themajority of the cases among older children, and husband's or mother'seducation was the most important variable in 21 of 26 cases. These new data,therefore, support the importance of parental education when controlling forother socioeconomic variables, but do not support the ealier finding thatmother's education had much stronger effects than husband's education.

In attempting to understand the channels through which educationoperates it is important to determine the extent to which its effects arethrough differences in the fertility behavior of women which in turn affectmortality and the extent to which it operates through diTerences in childcare practices unrelated to the biology of reproduction.-

Hobcraft et al (1983) and others have shown the importance of childspacing on the survival chances of the children. Other studies have shownthat family size, birth order and maternal age are important factors affectingchild survival. Does education have its effect only because more educatedwomen have fewer, better spaced children? Hobcraft et al (1984a) have estima-ted the effects of these biological determinants in a multivariate analysiswhich includes parental education and this allows us to determine if theeffects of parental education are eliminated when biological variables areincluded.

The Hobcraft et al study included data from 39 developing countriesusing WFS survey data. This paper used multivariate analysis to analyze childsurvival. The authors included the biological and demographic factors of sexof child, birth interval, age of mother and parity. Their results confirm thevery substantial importance of birth internal (child spacing) on childsurvival. Nevertheless controlling for all these biological factors they findthat parental education continues show substantial impacts of child survivalstatus.

Other Studies

There have been a substantial number of other studies published inthe past couple of years which provide either general insight on the educationand health relationship or provide empirical estimation of the relationship inone or several countries using household data. A conference held at theUniversity of Michigan in March 1981 entitled "Literary, Education and HealthDevelgyment" provides a collection of a wide variety of papers on the sub-ject.- Papers given at that conference by Caldwell and McDonald, Rosenzweigand Schultz, Simmons and Bernstein and Trussell and Preston and Chowdhury bprovide estimation of the relationship between education and child survival atthe household level.

The methodology of these studies vary enormously. The results showmore consistency. Caldwell and McDonald use cross tabulation of data andmultiple classification from ten WFS countries and conclude "Taking the ten

-4-

countries as a whole, the parents' education has over three times the impacton child mortality as father's occupation. Mother's education is somewhatmore important than father's education and the jump from primary schooling tosecondary schooling is twice as important as that of the original step toprimary schooling ..... further ..... establishes the fact that more educatedmothers achieve lower child mortality in spite of the considerable disadvantgeof earlier weaning" (Caldwell, 1982, 264).

Rosenzweig and Schultz used data from the 1973 Colombia census totest, not only the relationship between maternal education and child mortalityand fertility, but also to determine how the impact of access to health andfamily planning varied by the education of the mother. This paper thusaddresses the question raised earlier of whether public health facilitiescould substitute for maternal education. They find education is stronglyinversely related to child mortality ratio and fertility. Furthermore, theyfind in urban areas public health activities act as substitutes for maternaleducation. In rural areas the interaction of maternal education and access topublic health showed no significance either because the measurement of'access" is more difficult in rural areas or other mechanisms are at work.

George Simmons and Stan Bernstein used household survey data fromrural Northern India to study the determinants of child survival. Theyseparate their analysis for males and females in the neonatal and postneonatalperiod. Controlling for other factors a measure of the parents combinededucation shows no relationship to mortality for males or females in theneonatal period. In the postneonatal period it is found that mortality offemale children was significantly lower where the mother had att77ded schoolthan when neither parent or only the father had attended school.- For malechildren, the coefficient for mother's education is marginally significant atIn%. In evaluationg these findings the authors stress that the levels ofeducation are very low in this sample. In addition, it is necessary to bearin mind that for all levels of mortality they found sex differences inmortality determinants and found strong evidence that child survival wasassociated with whether or not a child of that sex was wanted. These findingssuggest that the determinants of mortality are more complex in thisenvironment than a simple model of survival maximization might suggest.

The Trussell and Preston paper presented at the 1982 MichiganConference was largely a methodological exercise to determine preferredmethods of estimating the magnitude of relationship between child survival andthe socioeconomic variables of residence, mother's education and husband'seducation and occupation. Using a variety of multivarite specifications theauthors found that the results were fairly robust with respect to methodology.and that mother's and father's education had proportionate and large effectson child mortality in Sri Lanka and Korea. They estimated that completion often years of schooling by a mother or father in Sri Lanka reduced childmortality by 50 to 55% while in Korea completion of high school reduced it by47%. (Trussell and Preston, 1982).

Another recent paper on the determinants of child mortality usesdata from three WFS surveys in Asia (Martin et al, 1983). The multivariatemodels used there show that education was second only to length of exposure torisk in determining child mortality. However, education's effects differed

substantially between countries. Mother's education had a stronger effect onmortality than father's education in the Philippines and Pakistan, but thereverse was true in Indonesia. Their finding confirm that education's effectin these countries is not just an "income" effect for mother's educationremains significant when father's education is included in the model and whereother measures of socioeconomic level are included in the one case where theywere available, the Philippines.

There have been a number of other published and unpublished studiesof the determinants of child survival in recent years using either WFS datasets (Meegama, 1980) or other large household surveys (Zachariah and Patel,1982). These surveys generally confirm the importance of maternal educationin multivariate models. A more complete review of these studies, theirmethodologies and the comparability of their findings is needed, but it iseven more important to begin to increasingly focus on the channels throughwhich education has its effects.

Frameworks for analyzing these channels are provided in work cur-rently being done. A Rockefeller/Ford Foundation joint conference held atBellagio in October 1983 provides numerous papers which directly or indirectlyaddress the issue of paternal education and child health.

Helen Ware's paper directly addresses the effect of maternal educa-tion on child mortality. Ware reviews various ways in which education wouldalter behavior which in turn reduces mortality. She discusses the design ofsurveys and survey instruments that might capture these effects. In theprocess of reviewing the various hypotheses she presents bits and pieces offascinating data on the linkages in various countries. This is probably thefirst article to attempt a partial review of the evidence on linkages and assuch is quite valuable. She also accurately summarizes why the linkages areof policy relevance. "Clearly, for example, if the relationship betweenmother's education and her child's survival is largely established by con-trolling for the birth weight of the child, then the policy implications arevery different from those that would arise from a proven link between maternaleducation, hygiene practices and child survival." (Ware, 1984, 210).

At the Bellagio conference papers by both Mosley and Chen andSchultz formulate models of child survival in which education's effects couldbe estimated. The two approaches differ substantially. In particular, thechannels through which education may operate in the Schultz's framework areprimarily two: those of the demand for health inputs and of the productionfunction for child survival. This economic model of the determinants ofhealth inputs is used by Schultz to discuss the channels (in terms of prices,preferences, technologies and opportunities) through which education mightaffect child survival. It is an interesting framework but the demands itmakes are so great that discretion often demands a second best solution(reduced form estimation) which ultimately surpresses the estimation of thechannels whereby education or other variables operate and only leaves us withestimates of the net effect of exogenous variables such as education.

The Mosley and Chen (1984) framework has five channels through whichany socio-economic variable can be expected to have their effects onsurvival. These are:

- 6 -

(1) Maternal factors - age, parity, birth interval.

(2) Environmental contaminations: air; food/water/fingers;skin/soil/inanimate object; insect sectors.

(3) Nutrient deficiency: calories, protein, micro nutrients (vitaminsand minerals).

(4) Injury: accidental, intentional.

(5) Personal illness control: personal preventive measures; medicaltreatment.

This framework provides guidance for studying the channels throughwhich education and other socio-economic variables can affect child health andsurvival. Unlike the proximate model of fertility which it resembles, it isnot possible to use this framework in a mathematical way to estimate theimpact of various socio-economic variables on child survival through thevarious channels. It is nonetheless a valuable framework for organizingfuture work on the link between socio-economic variables and child health.

The Bellagio paper by Tekce and Shorter represents the first attemptto use the Mosley-Chen framework to estimate both the impact of education onchild survival and to estimate some of the channels through which educationand other socio-economic variables have their effect on child survival. Thefirst part of their paper estimates the relationship between socio-economicfactors and child survival and the second part examine the intermediatemechanisms. Mother's education was found to have significant net effectson: the use of a trained birth attendant, DPT immunization, use ofprofessional health 8are when a child is ill and nutritional status of thesurviving children.' None of the other socio-economic variable wassignificantly related to more than 2 of the 5 intermediate variables. (Theonly variable not related to education was presence of soap in the toiletarea. This was affected, however, by housing quality, household income andoccupation of head of household). Thus this paper provides a rich analysis ofthe role of education in determining child survival in Amman, Jordan.

Conclusion

The recent publication of the Michigan and Bellagio Conferencepapers on child survival and the WFS survey results on child survival put usin a much better position to assess education's role on determining childsurvival than we were in 1980. The relationship between parental educationand child survival does not appear to be the spurious effect of education'sassociation with other socio-economic variables nor does it appear to resultexclusively from the relationship between education and reproductivebehavior. We are still uncertain, however, of what channels education doesact through and considerably more work needs to be done of the type done forJordan by Tekce and Shorter.

It will also be necessary, however, to do more work on modeldevelopment as well as data collection which will allows us to assess thequantitative impact of education on survival through different channels.

- 8 -

Footnotes

I/ The authors-of study from which Table 1 is taken attribute the deviationto probably reporting errors, but their argument for why there are moredeviations for husband's than mother's education is not completelyconvincing.

2/ If father's in the 7+ group are assumed to have on average more years ofschooling than mothers in that group, the difference in the magnitude of"'effect" between mother's and father's education is even greater.

3/ Mosley (1983) found that the strength of eduction's affect depend on thecommunity. Specifically Mosley found in poorer areas there were widereducation differentials.

4/ This distinction is important for designing policies for childsurvival. If the effects are biological, then policy requires the changein reproductive behavior while if they are child care, then it's possibleto reduce mortality without necessarily changing reproduction. Ideallyone would attempt to affect mortality through both channels if possible.

5/ These papers were published together in Health Policy and Education, Vol.2, No. 314 March 1982, edited by Robert N. Grosse and John Friedl.

6/ This latter was a very small sample and the results will not be discussedhere. It is a fairly unique study, however, since it is prospectiverather than retrospective. A review of the paper would be valuable foranyone planning future survey work.

7/ The effects of education remained strong when a control was introducedfor land ownership. Suggesting that education was not simply a proxy forwealth.

8/ Mother's education has a positive relationship with nutritional statusdespite the fact that more educated women were shown to breastfeed forshorter periods.

- 9 -

Bibliography

Arriaga, Eduardo E. (1979a), "Infant and Child Mortality in Selected LatinAmerican Countries," mimeo, Washington, D.C.: U.S. Bureau of Census.

, (1979b), "Infant and Child Mortality in Selected Asian Countries,"mimeo, Washington, D.C.: U.S. Bureau of the Census.

Behm, Hugo, (1979), "Socioeconomic Determinants of Mortality in LatinAmerica," Proceedings of the Meeting on Socioeconomic Determinants andConsequences of Mortality, Mexico City. Geneva: WHO.

Caldwell, John C. (1979), "Education as a Factor in Mortality Decline: AnExamination of Nigerian Data," Proceedings of the Meeting onSocioeconomic Determinants and Consequences of Mortality, Mexico City.Geneva: WHO.

Caldwell, John and Peter McDonald (1982), "Influence of Maternal Education onInfant and Child Mortality: Levels and Causes," Health Policy andEducation (2).

Chowdhury, A.K.M. Alauddin, (1982) "Education and Infant Survival in RuralBangladesh, "Health Policy and Education (2).

Cochrane, Susan, Donald, J. O'Hara and Joanne Leslie, (1980,), "The Effects ofEducation on Health," World Bank Staff Working Paper No. 405, July 1980.

Hobcraft, John, John W. McDonald and Shea Rutstein, (1983), "Child SpacingEffects on Infant and Child Mortality," Population Index (49, 4) Winter,1983.

7 (1984a), "Demographic Determinants of Infant and Early ChildhoodMortality: A Comparative Analysis," mimeo World Fertility Survey(International Statistical Institute) June 1984.

, (1984b), "Socioeconomic Factors in Infant and Child Mortality: A CrossNational Comparison," Population Studies (38, 2) July 1984.

Martin, Linda el at (1983), "Covariables of Child Mortality in thePhilippines, Indonesia and Pakistan: An Analysis Based on Hazard Model,"Population Studies (37) November, 1983.

Meegama, S.A. (1980) "Socioeconomic Determinants of Infant and Child Mortalityin Sri Lanka: An analysis of Post-War Experience". InternationalStatistical Institute, World Fertility Survey, Scientific Report No. 8,London.

Mosley, W. Henry, (1983), "Will Primary Health Care Reduce Infant and ChildMortality? A Critique of Some Current Strategies with Special Referenceto Africa and Asia," Paper presented in IUSSP Seminar on Social Policy,Health Policy and Mortality Prospects, Paris. Paris: Institute Nationald'Etudes Demographic.

- 10 -

Mosley, W. Henry and Lincoln C. Chen (1984), "An Analytical Framework for theStudy of Child Survival in Developing Countries," in Child Survival:Strategies for Research W. Henry Mosely and Lincoln C. Chen (editors)Population and Development Review (a Supplement to Volume 10).

Rosenzweig, Mark R. and T. Paul Schultz (1982), "Child Mortality inColombia: Individual and Community Effects," Health Policy and Education(2).

Schultz, T. Paul (1979), "Interaction of Relations Among Mortality, Economicsof the Household and the Health Environment," Proceedings of the Meetingon Socioeconomic Determinants and Consequences of Mortality, MexicoCity. Geneva: WHO.

(1984), "Studying the Impact of Household Economic and CommunityVariables on Child Mortality," in Child Survival: Strategies forResearch W. Henry Mosley and Lincoln C. Chen (Editors), Population andDevelopment Review (a supplement to Volume 10).

Simmons, George B. and Stan Bernstien (1982), "The Educational Status ofParents and Infant and Child Mortality in Rural North India," HealthPolicy and Education (2).

Tekce, Belgin and Frederic C. Shorter, (1984), "Determinants of ChildMortality: A Study of Squatter Settlements in Jordan," in ChildSurvival: Strategies for Research W. Henry Mosley and Lincoln C. Chen(editors), Population and Development Review (a Supplement to Volume 10).

Trussell, James and Samuel Preston (1982), "Estimating the Covariates ofChildhood Mortality from Retrospective Reports of Mother," Health Policy andEducation (3).

Zachariah K.C. and Sulekha Patel, (1982), "Trends and Determinants of Infantand Child Mortality in Kerala" Population and Human Resources DivisionDiscussion Paper No. 82-2, January 1982 (World Bank).

Ware, Helen, (1984), "Effects of Maternal Education, Women's Roles and ChildCare on Child Mortality," in Child Survival: Strategies for ResearchW. Henry Mosley and Lincoln C. Chen (Editors) Population and DevelopmentReview (a Supplement to Volume 10).

Table 1: UNDER FIVE MORTALITY RATES (PER 1,000 BIRTHS)BY PARENTAL EDUCATION

/aMother's Education Husband's Education

0 1-3 4-6 7+ 0 1-3 4-6 7+ Total

AfricaSenegal 296 (178) (113) (19) 310 (261) (138) (74) 287Lesotho (224) 215 185 169 185 212 196 169 190Kenya 181 164 128 111 185 208 159 127 162Sudan 146 ------- 109 ------- 155 152 104 84 140

AmericaHaiti 226 (176) (206) (89) 230 204 (191) (150) 214Peru 237 171 98 55 241 217 150 75 170Dominican Republic 198 140 122 82 175 134 133 112 139Mexico 153 118 87 50 155 121 96 53 114Columbia 146 127 82 49 139 129 86 68 112Costa Rica 157 100 85 45 138 113 81 44 95Paraguay 110 88 67 32 (87) 88 76 41 77Guyana (83) (89) 80 64 (38) (89) 87 66 71Panama 134 90 52 43 100 91 60 45 65Venezuela 79 60 60 35 84 61 58 43 61Jamaica (82) (78) 74 51 (57) (69) 70 55 58Trinidad & Tobago (74) (67) 59 43 80 117 57 43 49

AsiaNepal 261 (204) (157) (136) 270 (216) (234) 167 259Bangladesh 222 198 186 (122) 230 221 191 176 215Pakistan 208 (143) (138) (112) 215 (172) 208 155 202Indonesia 193 194 143 77 199 205 165 99 180Thailand 145 105 110 (38) 151 (162) 109 83 116Jordan 112 83 84 67 112 115 105 77 100Syria 104 (75) 75 50 109 94 86 70 95Philippines 130 118 94 53 118 122 96 62 90Sri Lanka 104 97 80 53 122 98 93 57 84Korea 107 94 74 56 102 (134) 90 67 83Malaysia 67 64 56 18 88 59 60 40 61Fiji 70 67 60 46 59 66 65 55 59

Average 152 122 102 67 148 140 116 84 127

/a These are current husbands of the mothers and may not be the father of the childrenwho may have been born in a previous marriage.

Source: From Table 3 Hobcraft, McDonald and Rutstein (1984b).

- 13 -

ThE EFFECTS OF EDUCATION AND URBANIZATION ON FERTILITY*

* Copyright 1983 Academic Press. Reprinted with Permission.

- 15 -

EFFECTS OF EDUCATION AND URBANIZATION ON FERTILITY

The purpose of this paper is to document the effects of education andresidence on fertility. The effect of education has been much morecarefully documented (Cochrane, 1979) than that of residence; however,both raise similar methodological issues.

First, neither education nor residence can affect fertilitydirectly in the same way as age at marriage, lactation, andcontraceptive use, but must operate through such variables. Althoughthese intervening channels are not the focus of this paper, they mustbe recognized if the utruew effects of education and residence are tobe assessed. If all these channels were included in the analysis, themeasured effect of education would be reduced to zero.

In addition to these problems of causal modeling, there is asecond problem--linearity. If the relationship between two variablesis linear, one can aggregate with impunity. Unfortunately, theevidence indicates that the relationship between education andfertility is nonlinear, and is in fact not even monotonic. In aprevious literature review, uniformly inverse relationships were foundin only 49 percent of the cross-tabular studies; significant negativecoefficients were found in only 31 percent of the regression studiesin which income and wealth were controlled along with age andresidence (Cochrane, 1979). In most cases, if the relationship wasnot monotonically inverse, fertility first rose with education andthen fell. In these cases, peak fertility was generally found atlower primary education. The new data reviewed in this paper confirmthe existence of two different patterns--the inverse pattern and aninverted 'U." Similarly, it is unlikely that the relationship betweenresidence and fertility is linear, although this has not beendocumented. Therefore, studies should probably be restricted to theindividual rather than the aggregate level.

The methodological issues involved in assessing the effects ofeducation and residence on fertility are discussed in detail below.

- 16 -

Following this discussion is a summary review of the literature onthese effects.

MEASURING THE EFFECTS OF EDUCATION AND RESIDENCEON FERTILITY: METHODOLOGICAL ISSUES

Measuring Education's Effect

To measure the effect of education on fertility, one must firsteliminate spurious correlations between education and other fertilitydeterminants, most obviously age and residence. In developedcountries, younger women are generally more educated than older women,and of course age is highly correlated with number of children everborn. Therefore, if age is not controlled, the effect of education onfertility will be overestimated. Likewise, education tends to behigher in urban than in rural areas, and fertility is generally higherin rural than urban areas. Thus if residence has effects other thanthose through education and if it is not controlled, education'seffect will be overestimated. Beyond these generalizations, however,there is not even agreement on how to control appropriately forresidence or exposure to childbearing. Moreover, although otherbackground variables such as caste probably determine both educationand fertility and therefore need to be controlled, these factors varyamong countries so that generalizations cannot can be made.

There is also considerable disagreement about which effect ofeducation one wishes to estimate, depending on the objective of theresearch. This paper focuses on the total effect of education. Asnoted above, education acts through many other variables, such as ageat marriage, lactation, and contraceptive use. If these interveningvariables (IV) are included in the analysis, then education's totaleffect on fertility will be biased. Therefore, education's effect onfertility through these variables must be added to its 'direct'effect. Table 1 lists a number of these variables; the effect ofeducation on each; the effect of the IV on fertility; and theresultant bias in estimating education's total effect if the IV iscontrolled but the effect of education on that IV is not added.These variables have been arranged under the three broad determinantsof fertility: the supply of children, the demand for children, andfertility regulation.

In most cases (14 of 15), the inclusion of an intervening variablewill bias the education variable positively (i.e., toward zero if theeffect is negative). On the other hand, while the direction of thebias can be determined in most instances, the magnitude can not, sinceit depends on how strong an effect education has on the interveningvariable. For example, if mass media exposure includes mainly writtencommunication, education will have a strong effect on exposure;inclusion of exposure in the analysis will then lead to a seriousunderestimate of education's effect (assuming that mass media exposurehas a strong effect on fertility). On the other hand, if mass media

- 17 -

TABLE 1 Education--the Intervening Variables Through Which It AffectsFertility, and the Biases Introduced by Including these Variables withEducation in Regression Models

Bias in EstimatingEducation's Effect

Effect of Effect of if Controlling forIntervening Education IV on IV Without AddingVariable (IV) on IV Fertility Indirect Effects

Supply of BirthsAge of marriage + - +BreastfeedingAbstinence

(noncontraceptive) - -

Demand for ChildrenIncomea + ? ?Wealth + ? ?Female wage + - +mass media + _ +Female labor participationa + ? ?Child labor participationa - + +Child schoolinga + - +Child survivala + - +Perceived cost of children + - +Perceived benefits ofchildren + +

Migrationb + - +

Ferti'lity RegulationContraceptive use + - +Knowledge of birth control + - +Access to birth control + - +

a Variables affected by as well as affecting fertility. Income isincluded in the list since it depends on labor participation, whichin turn is affected by fertility; income is also affected byeducation independently of labor participation as a result ofassortive mating.

b Education may affect migration, but migration's effect on fertilityis uncertain. Most of its effect is probably captured by residence.

- 18 -

exposure represents primarily radio, the bias resulting from inclusionof this variable will be less serious.

These points can be illustrated by Mason and Palan's (1980)Malaysian study. Starting with the basic controls of age and maritalduration,2 they estimate the coefficient of education for sixracial-residence groups and then successively show the effect on theseestimates of the following variables: occupation prior to marriage,residence when growing up, husband's education, a sanitary andamenities scale, family income, and the minimum education for sons.When these variables are added, significant inverse (linear ornonlinear) relationships are eliminated in four out of five groups; inthe fifth group--rural Malays--for whom the positive effects ofeducation were initially strongest, the negative effect of educationis increased. (In the sixth group, the coefficient was notsignificant initially.) The neutralizing effect of the variables

differs from group to group. This variation among groups in thechannels through which education operates makes it impossible toformulate a priori statements about the size of the biases introduced(or eliminated) by introducing variables.3

The indirect effects of education can be easily adjusted for ifsufficient information is available and causation is unidirectional.However, if causation runs in two directions, it is much moredifficult both conceptually and empirically to determine the effect ofeducation on fertility. Two kinds of problems may arise here. First,education and some other variable may be simultaneously determined.The most important example is the education of husband and wife.Assortive mating will result in more educated women being married tomore educated men. To the extent that the husband's education alsoaffects fertility, it would be a mistake to attribute all ofeducation's effect to either the wife or the husband. If the wife'seducation is used alone, its effect will be biased away from zero; ifboth the husband's and the wife's education are included, then theeffect of each education measure will be biased toward zero. Althoughthere is no clear-cut solution4 to this problem, it helps to beaware of the upper and lower limits on 'education's" effect (seeHermalin and Mason, 1980).

A second kind of problem arises if fertility, rather thaneducation, is jointly determined with some other variable. Forexample, if fertility and female labor participation areinterdependent, then the inclusion of the latter causes thecoefficients of other explanatory variables such as education to bebiased. Since the importance of this bias is a matter of degree,there is considerable disagreement over which variables should betreated as simultaneous with fertility.5

In Table 1, those variables most commonly considered to besimultaneous with respect to fertility have been marked. There are novariables on this list which might be considered to have feedbackeffects on education, primarily because these variables are measuredin adulthood, generally long after education has already beencompleted. The one important exception might be age at marriage ifthere is overlap between marriage ages and levels of education.

- 19 -

To summarize, residence and age are necessary controls if one isto avoid overestimating the negative effect of education.6

Inclusion of other variables determined by education may bias theeffect of education (usually toward zero). Inclusion of husband's andwife's education together will reduce the measured effect of eachindividual measure, while exclusion will lead to an overestimate(unless they differ in the sign of their effect); the "proper'specification will depend on the use to be made of the model.Finally, inclusion of simultaneous variables will cause biases ofunknown direction and magnitude.

Measuring Residence's Effect

The effect of education on fertility cannot be studied separately fromthat of residence. The reason for this is that many of the factorsaffecting individual decision making are determined at least in partby the community of residence. These factors include presence in thecommunity of contraceptive services; schooling opportunities andhealth facilities; economic opportunities and costs, such as thedemand for labor of men, women, and children; the costs of food andhousing; and the less easily defined factors of social milieu,climate, and exposure to disease. All of these factors vary betweenurban and rural areas. Ideally, one would control for these factorsindividually; however, sufficient data are rarely available.Therefore, residence is used as a control to approximate thesefactors. This is at best a very gross control, especially if adichotomous variable is used for residence.

The effect of residence on fertility operates through all theseintervening factors. As in the case of education, if these factorsare included with residence in a model of fertility determination, abias will be introduced in estimating residence's effect. Table 2shows the direction of such biases. It should be recognized thatbecause the intervening community variables (ICV) will in turn affectintervening family variables (IFV), the inclusion of the latter willaffect estimates of the effect of residence. One major advantage ofICVs over IFVs in an analysis is that, unlike the ICVs, the IFVs maybe simultaneously determined with family fertility. Thus althoughboth introduce biases, the ICVs are exogenous to the family, and thebiases are of known direction.

Interactions, Nonlinearities, and Measurement Problems

To this point, residence and education have been discussedseparately. However, it is important to recognize that interactionsmay exist. These interactions can arise from (1) the effects of onevariable on the other; (2) the fact that education operatesdifferently in various environments; and (3) a combination ofnonlinear effects and different levels of education in different areas.

The first of these possibilities depends on whether place of birth

- 20 -

TABLE 2 Urbanization--the Intervening Variables Through Which ItAffects Fertility, and the Biases Introduced by Including theseVariables with Urbanization in Regression Models

Bias in EstimatingEffect Residence's Effectof Urban- Effect of if Controlling forization ICV or ICV or IFV Without

Intervening on ICV IFV on Adding IndirectVariable or IFV Fertility Effects

Intervening Community Variables (ICV)

Presence of familyplanning facilities + - +

Presence of schools + - +Presence of health

facilities + - +Demand for labor of

Men + + -

Women + +Children - + +

Housing costs + - +Food costs + - +Sociocultural milieu + - +Disease exposure ? - ?

Intervening Family Variables (IFV)Incomea + ? ?Female wage + - +Mass media + - +

Knowledge of birth control + - +Access to birth control + - +Female labor participationa + ? ?Child labor participationa - + +Child schoolinga + - +Child survivala + - +Cost of children + - +Benefits of children - + +Modernism + - +

aEndogenous variable.

- 21 -

or place of current residence is used as the residence variable. Inthe former case, the places of birth, childhood, and adolescence canaffect educational opportunities and thus determine educationalachievement; thus the total effect of place of birth on fertilitywould have to-include its effect through educational achievement. Ifcurrent residence is used, education may affect residence throughmigration; thus education's effect would have to include its effectthrough migration. (Migration is included in Table 1. However, thereis little evidence on whether migration per se or residence before andafter migration affects fertility, although those who migrate fromrural to urban areas generally have lower fertility.)

The second of these possible interactions, the fact that educationoperates differently in urban and rural areas, has frequently beendiscussed, but with little certainty. One possible interaction isthrough labor markets. Increased education opens modern-sectoremployment opportunities for women in urban areas; it does notgenerally have this effect in rural areas. Of course, migration wouldallow rural women to avail themselves of urban opportunities, but athigher cost than that incurred by urban women.

Although such behavioral interactions between education andresidence are not precisely defined, a statistical interaction clearlyexists. The evidence that urban areas have a different educationaldistribution from rural areas is very convincing. In addition, it isfairly clear that education's effect on fertility is nonlinear in someenvironments. Therefore, if linear estimates of education's effectare made, that effect will appear different in urban and rural areas;if the correct nonlinear specification is used and no otherinteractions exist, that effect will not differ between urban andrural areas.

In addition to these considerations of interaction andnonlinearity, there are problems of measurement. Both education andresidence can be measured as either dichotomous or continuousvariables. It is much more common, however, to use an urban/ruraldichotomy for residence than to use a literate/nonlilterate dichotomyfor education. Therefore, the measurement of education is usuallymore detailed than that of residence. This point will be raised inthe following review of the literature on the effects of education andresidence on fertility.

LITERATURE REIEW

As noted earlier, only studies having a direct or indirect age controlwill be included in the present review. This review is also limitedto studies having a minimum sample size or degrees of freedom of over150 in the relevant age-residence group. Moreover, because of thenonlinearities noted above, studies based on regional aggregates ofdata will not be included. Finally, if the same methodology were usedin all studies, the results would be easy to summarize; however,studies differ substantially in how they analyze fertilityideterminants. Therefore, this review will be restricted to studies in

- 22 -

which the dependent variable is a measure of completed fertility,either children ever born or the total fertility rate, and in whichage or marital duration is implicitly or explicitly used as a control.

The subsections below first review a study that uses a commonmethodology to estimate the effects of education and residence onfertility for several countries. This study, prepared by Rodriguezand Cleland for the 1980 World Fertility Survey (WFS) Conference inLondon, addresses the separate effects of education and residence, aswell as their interaction. Next, two other types of studies ofeducation's effect on fertility are examined: Hermalin and Mason'smeta analysis, and a number of multivariate studies within countries.Finally, two important considerations in studying the effects ofeducation and residence on fertility are examined: the issue of thestability of educational differentials over time, and evidence onurban-rural differences in fertility.

Comparative Analysis of WFS Data

The Rodriguez and Cleland paper (1980) is a comprehensive report onthe effects of education, residence and a number of other explanatoryvariables on fertility for 22 countries in which WFS surveys have beenconducted. This paper is extremely useful in illustrating the pointsdiscussed above, as well as in quantifying the relationships betweeneducation and residence and fertility.

Rodriguez and Cleland use regression analysis both to create theirprimary dependent variable--marital fertility--and to estimate theeffects of education and residence on fertility. The dependentvariable is constructed by estimating the effect of marital durationand duration squared on fertility in the last 5 years (weighted byexposure), and then predicting total fertility over 25 years ofmarriage using the estimated effects of duration. Thus the dependentvariable, an approximation of completed fertility based on recentfertility, resembles the total fertility rate in itscharacteristics.7

Education

Rodriguez and Cleland estimate the effects of education and othervariables on fertility using dummy variables and regression. Averagemarital fertility can be determined for every category of education(see Table 3). Other variables can also be introduced, and averagemarital fertility determined controlling for these factors.

Table 3 shows the relationship between fertility and the educationof the wife (or woman) and that of the husband. In Panel A, fertilityis estimated, using total fertility rates, for all women in thehousehold; in Panel B, only ever-married women are included, andfertility is estimated at marital durations of 25 years. Beforereviewing the data in Panel B, it is important to determine ifrestricting the sample to ever-married women causes any bias. If

TABLE 3 Effect of Education on Predicted Completed Fertility

Panel A. Panel B.Total Fertility Rate (all women) Adjusted for Marital Duration (ever-_arried women)

Woman's Education Wife's Education Husband's Education

No Lower Upper No Lower Upper No Lower UpperCountry School Primary Primary Secondary School Primary Primary Secondary School Primary Primary Secondary

Bangladesh 6.2 6.5 6.9 5.0 6.2 6.6 7.1 6.1 6.0 6.9 6.6 6.8Fiji -- -- -- -- 4.8 5.2 4.8 4.1 4.9 5.3 4.7 4.3Indonesia - -- -- -- 4.9 5.5 5.6 5.4 4.5 5.4 5.3 5.7Korea 5.8 5.2 4.5 3.3 6.2 5.5 5.0 4.0 6.1 6.1 5.5 4.5Malaysia 5.2 5.1 4.6 3.2 6.3 6.0 5.4 4.6 5.9 6.3 5.8 5.1Nepal -- -- -- -- 6.1 6.5 6.1 2.8 6.1 5.4 6.2 5.4Pakistan 6.6 (5.9) (5.7) 3.6 7.0 6.7 7.1 5.8 7.0 6.6 7.3 6.8Philippines 5.2 6.9 6.0 3.8 6.4 7.1 6.6 5.3 6.5 7.2 6.6 5.5Sri Lanka -- -- -- -- 5.3 5.1 5.3 4.8 5.4 5.3 5.2 5.0Thailand -- -- -- -- 5.5 5.7 5.3 3.6 5.1 6.8 5.5 4.0

Colombia 6.4 5.5 3.5 2.5 6.7 5.9 4.1 3.3 6.7 6.1 4.3 3.3 1Costa Rica (4.9) 4.5 3.3 2.7 5.2 4.5 3.4 3.3 4.7 4.5 3.6 3.3Dominican Republic (6.6) 6.4 4.1 (2.3) 6.9 6.7 4.9 3.7 6.6 7.1 5.1 3.5 rNGuyana -- 5.4 5.3 4.0 6.0 5.1 4.8 4.4 4.6 5.3 5.0 3.9Jamaica -- 5.4 5.1 3.2 6.6 5.3 5.0 3.4 6.0 5.4 4.9 3.4Mexico 7.6 6.6 4.7 3.3 7.4 7.2 5.5 4.6 7.6 7.2 5.9 4.7Panama - 6.1 4.3 2.9 6.7 6.2 4.5 3.6 6.6 6.2 4.5 3.7Peru 7.0 6.4 3.7 3.4 7.7 6.9 5.4 4.5 7.2 7.4 6.3 5.1

Jordan 9.2 6.8 5.4 (3.6) 9.3 7.7 6.7 5.6 9.3 9.1 7.9 6.8Kenya 8.4 8.9 8.0 5.5 7.4 8.1 7.7 7.8 7.2 8.2 7.9 7.8

Average of all

available countries 6.6 6.1 5.0 3.5 6.4 6.2 5.5 4.5 6.2 6.4 5.7 4.9

Average forcountries with datain Panel A only 6&6 6.1 5.0 3.5 6.8 6.4 5.6 4.7 6.5 6.6 5.8 5.0

Note: Dash indicates not available or n less than 2501 parentheses denote n less than 500.

Source: Rodriguez and Cleland (1980:Tablea 4, 7, A-2).

- 24 -

education has an effect on age at marriage, as many studies haveshown, then this restriction, used in most fertility surveys, willresult in an underestimation of the effect of education on fertility.Panel A offers some guidance in making this determination because itincludes education and total fertility data for married and unmarriedwomen combined. On average, the difference between the effects ofeducation shown for all women and only married women are substantial(compare Panel A with the effects of wife's education in Panel B).Over the entire education range, the difference in fertility is 3.1when all women are included, but only 2.1 if only married women areincluded.8 Thus, a priori, surveys of married women will tend tounderestimate the effect of education on fertility substantially,depending on the importance of education in determining age atmarriage. This point needs to be considered in reviewing anycomparative analysis of data drawn from surveys of ever-married woraen.

These data for ever-married women show that the relationshipbetween education and fertility is monotonically inverse for women inLatin America; outside Latin America, female education ismonotonically related to fertility only in Jordan, Korea, andMalaysia. (Sri Lanka is a marginal case.) For males, therelationship is monotonic only in five of the eight Latin Americancountries, Jordan, and Sri Lanka.

This irregular relationship between female education and fertilityoutside Latin America is not inconsistent with the pattern found inCochrane (1979). The nonlinearities differ according to the educationlevel at which the highest fertility is observed, the regularity ofthe increases and decreases in fertility on either side of the peak,and the magnitude of the differences in fertility am.ong those with noeducation and those with maximum education. In Sri Lanka andPakistan, the overall pattern is irregular, showing two peaks; in theother countries, only one peak is evident, and the fertilitydifferences between the extreme education categories are typically atleast one-half a child. Although these irregularities defy simpledescription, they are sufficiently persistent and based on adequatelylarge samples that they cannot be dismissed as mere flukes.

In an attempt to determine if such nonlinearities are regionallydetermined or a function of development, some additional analysis wasdone for this paper: literacy, urbanization, caloric intake, income,and total fertility rates were compared for countries with and withoutmnonotonic fertility-education relationships. Simple cormparisons showthat only for per capita income is there a clear separation betweenthe two groups of countries: all countries with 1978 per capitaincome at or below $510 have a pattern of increased fertility witheducation prior to a decline; all countries with an income of $740 orabove have monotonic inverse relationships.

However, income is not the only important variable for some verylow-income countries, such as Sri Lanka, that have small increases infertility, or for richer low-income countries, like the Philippines,that have large increases in fertility before a decline is observed.A test of the relative importance of variables in explaining the shapeof the educationfertility relationship is shown in Table 4. The

TABLE 4 Relationship Between Fertility at Different Education Levels and Aggregate Characteristics

Fip O FEJP-P--lo FF%o 1UP-FLP FS-VLP 1

S- 1P

Wife's Education

GNP -. 002 (3.99) -- -.001 (2.60) -- --

TPR - -- ---

Urbanization -.020 (1.98) -.044 (5.53) -.047 (2.31) -.025 (2.95) -- --Literacy -- - -- --Daily calories -- -.002 (2.83) -- -.002 (2.34) -- -_

Conltant -.0495 -3.43 -6.82 -2.94 -- --R .39 .88 .38 .67 -- --d.f. 12 12 12 12 -- -

Husband's Education

GNP -- -.001 (2.92) -- -.001 (3.24) --TFR - .254 (2.03) -- -- --Urbanization -. 023 (2.39) -. 043 (5.57) -.054 (3.77) -.020 (2.95) -.031 (2.35) --Literary .022 (2.25) .017 (2.08) - -- -- -Daily Calories -- .002 (2.90) -- .003 (4.63) .003 (2.77) -_

Cons2tant .555 -4.49 -4.62 -4.99 -- --R .38 .83 .71 .77 .66d.f. 13 12 12 12 12 -- in

Notes m e dependent variable is the difference in fertility between education groups: those withlower prisary school and no education (FLrp No, upper primary and no education (FupPNho,etc. Only significant coefficients are showng t values are in parentheses.

Sources Regressions based on data in Table 3 and World Bank (1980).

- 26 -

differences in fertility between particular levels of education wereregressed on per capita income, the total fertility rate, urbanization(proportion urban), the literacy rate, and per capita calories.Urbanization is the most consistent variable explaining thedifferences in fertility between education groups. For the first foursteps of education given in Table 4, the greater the degree ofurbanization, the more likely it is that fertility will fall with eachincrement in education, and the larger these decreases will be. GNPbehaves similarly for the differences between no schooling and upperprimary, and between lower primary and upper primary. Caloric intakeis also significant at these steps in education. Rather surprisingly,neither aggregate education (literacy) nor total fertility rate issignificant in explaining the changes in fertility between particularlevels of education. This suggests an important interaction betweenpoverty (GNP and caloric intake) and urbanization and the effect ofeducation on fertility. It is interesting to note that, for husband'seducation, urbanization and GNP have the same relationship as forwife's education; however, caloric intake, literacy, and totalfertility rate have perverse signs. Of these, caloric intake is themost interesting in that it is significant in three equations.

In Cochrane (1979), it was shown that female education was muchmore likely than male education to be inversely related to fertility.This is confirmed in the WFS data by the unweighted average fertilityfor males and females in each education group. These average values,given at the bottom of Table 3, show that there is a difference of 1.9children over the education spectrum from no education to secondaryschooling for females, and a difference of only 1.3 for males. Inaddition, males with some education have higher fertility on averagethan those with no schooling--6.4 versus 6.2 (these values arereversed for females). In considering the extent to which thesedifferences accurately reflect the effect of education, severaladjustments need to be considered. Adjustments for residence and forspouse's characteristics are addressed by Rodriguez and Cleland, andare discussed in the subsections below.

Residence

Table 5 summarizes Rodriguez and Cleland's urban-rural fertilityestimates. The difference between urban and rural areas is greaterfor the data on all women (1.6 children) than for those on marriedwomen only (1.2). This is similar to the finding for education.However, the residential differential using either group of women issmaller than the educational differential, perhaps because residenceis measured dichotomously, whereas education is classified into fourgroups.

A particularly interesting finding is that urban fertility isalways lower than rural fertility when all women are included, but notwhen only married women are sampled. In Bangladesh, Indonesia, andPakistan, the marital fertility of urban women is higher than that ofrural women. In Indonesia and Pakistan, this difference is large,

- 27 -

TABLE 5 Effect of Residence on Predicted Completed Fertility

Total Fertility Marital Duration Adjusted(all women) (ever-married women)

Country Rural Urban Rural Urban

Bangladesh 6.2 6.0 6.3 6.5Fiji 4.6 3.6 5.2 4.3Indonesia 4.8 4.6 5.1 5.6Korea 5.1 3.7 5.9 4.5Malaysia -- -- 6.1 5.5Nepal 6.2 -- -- --

Pakistan 6.4 6.2 6.8 7.3Philippines 6.0 3.9 6.8 5.4Sri Lanka 3.8 3.2 5.3 4.8Thailand 4.9 2.9 5.5 4.3

Colombia 6.6 3.5 7.2 4.2Costa Rica 4.7 2.9 4.8 3.3Dominican Republic 7.1 4.1 7.6 5.0Guyana 5.1 4.3 5.1 4.1Jamaica 5.3 3.9 5.4 4.1Mexico 7.3 4.8 7.7 5.7Panama 5.7 3.3 5.9 3.9Peru 7.0 4.6 7.6 5.8

Jordan 9.8 7.1 9.7 8.5Kenya 8.4 6.1 8.0 6.5

Total 6.0 4.4 6.4 5.2

Note: Dash indicates not available.

Source: Rodriguez and Cleland (1980).

one-half a child. This implies that, particularly in these twocountries, age at marriage is very important in explaining residentialdifferentials.

Multivariate comparisons of urban-rural differences in fertility,similar to those in Table 4 and using the same variables, show thatonly the degree of urbanization is significant. The higher the levelof urbanization in a country, the greater the difference between ruraland urban fertility. Findley and Orr (1978) also found the percentageof the population urban significant in explaining urban-ruraldifferentials in fertility.9 These findings suggest that in

- 28 -

countries with greater degrees of urbanization, those classified asurban are more likely to be from large cities where fertility is infact lower than in rural areas.

Interaction of Residence and Education

The question of interaction between residence and education was alsoaddressed by Rodriguez and Cleland. Although they found that suchinteractions were not statistically significant in most instances,their test of interaction was not symmetrical for husband's and wife'seducation.

Husband's education was adjusted for residence by the inclusion inthe regression of a dummy variable for residence. In many cases,there was no perceptible difference in education's effect before andafter this adjustment. In those cases where the interaction term wassignificant, the residential adjustment reduced the impact ofhusband's education in all cases except Kenya, where education'seffect became more positive, and in Guyana, where there was nochange. Overall, the average (unweighted differential) beforeadjustment was -1.9; after adjustment the average was -0.85.

On the other hand, wife's education was adjusted not only forresidence, but also for husband's characteristics (education,occupation, and work status).10 These adjustments reduced femaleeducation's average effect from -1.9 to -0.955, making it lessnegative in most cases, although in one case (Indonesia), it made thateffect negative rather than positive. Adjustment for residence seemsdesirable to estimate the "true effect' of education; however,adjustment for husband's work status and occupation is not desirablesince these are at least partially the effect of the wife's educationthrough assortive mating.

Residence's effect was also measured by Rodriguez and Cleland,adjusting for all other variables (husband's education, occupation,and work status, and wife's education and work status). Again, suchadjustments generally reduced the effect of residence from an averageof -1.2 to -0.66.

The effects of these various adjustments are summarized in Table6. After controlling for other variables (some of which areendogenous to education, residence, or both), the effect of femaleeducation is still to reduce fertility by almost one child; the effectof male education is one-third as large; and the effect of urbanresidence is about two-thirds of a child. Although these figuressupport a strong negative effect of female education, this effect isnot linear, as was noted above. After all other variables have beencontrolled, the difference in fertility for married women is only atenth of a child between no and some primary education, and thenegative effect increases thereafter. In contrast, the fertility ofhusbands with some primary education is one-third of a child largerthan that of husbands with no education after controlling for allother factors, including residence, wife's education, and work

- 29 -

TABLE 6 Differences in Predicted Completed fertility Between Specific Education-ResidenceGroups

Wife's Education Husband's Education Residence

F1WFFNO ruP-rs tN0 ps t F1,PFN0 FUP -FNo pSmruO LtU-tR

All women -0.5 -1.6 -3.1 -- - -- -1.6Married womn -0.2 -0.9 -1.9 0.0 -0.5 -1.3 -1.2

Controlling forresidence -- -- -- 0.6 -0.3 -0.9 --

Plus husband'scharacteristics -0.9 -0.4 -1.0 - -- -- --

Plus all othervariables -0.1 -0.4 -0.9 0.3 0.0 -0.3 -0.7

Sources Sumarixed from Rodrigues and Cleland (1980) using umnwighted averaging of results.

status. Thus adjustment for other variables does not eliminate thenonlinear relationship between education and fertility.

Other Studies of Education's Relationship to Fertility

Meta Analysis

Another study using WFS data from several countries was done byHermalin and Mason (1980). This paper, which addressed strategies forcomparative analysis of WFS data, used an approach falling betweenmicro and macro analysis. Instead of using individual or countrywidedata to estimate regressions between education and fertility, it usedas the dependent variable the average fertility of women of variouseducation and marital duration groups from WFS tabulations. Theindependent variables were education and marital duration, and theirvalues were estimated by the approximate midpoints for each group.

Using these data points from WFS tabulations in each of 10countries, Hermalin and Mason fitted a nonlinear model to the data.The model used was ln (P + .5) - I + bE + c ln D, where P is parity, Dis duration, and E is education.1 This formulation implies thatthe higher the level of education, the stronger its effect. In allcases except Nepal, the coefficients for education were negative,although they varied in size from -0.0077 in Sri Lanka to -0.04085 inPanama and +0.02531 in Nepal. These variations in the effect ofeducation were then studied at the national level to determine whatmacro characteristics affect the relationship between education andfertility.

The analysis most relevant to this review is that which relatesthe effect of education on fertility in a country, B1 , to thatcountry's level of development and to characteristics of the educationsystem. Generally, the more "modernw the country, the more negativewas education's effect. Five correlations were significant at the 5

- 30 -

percent level for a two-tailed test: (1) 1957/58 per capitaexpenditures on education, in U.S. dollars; (2) percentage rural; (3)females in teacher training as a percentage of all students in teachertraining, secondary and higher, 1960; (4) percentage of economicallyactive females aged 14 to 65 in occupations other than agriculture inthe early 1960s; and (5) females as a percentage of total schoolenrollment at first, second, and third levels. The most interestingfinding is the robustness of the effect of the percentage rural:

We would conclude on the basis of the analysis thus far that theeffect of education on fertility is weaker in more ruralsettings. Whether this is due to the content of education or tothe opportunity of individuals to use education remains to bestudied and the general hypothesis needs to be further tested withmore observations (Hermalin and Mason, 1980:116).

The summary of Hermalin and Mason's work presented here is veryabbreviated. However, it does illustrate the pattern of nonlinearityin the education-fertility relationship, as well as the interaction ofeducation and residence.

Multivariate Studies Within Countries

A great number of multivariate studies of fertility determinants havebeen done on micro data within countries. For this reason, thepresent discussion must be selective. For comparability, only studiesusing children ever born as the dependent variable and a sample sizeof over 150 will be included. Estimates of both linear and nonlinearrelationships between education and fertility will be examined.

The Appendix Table summarizes the results of 30 multivariatestudies of the effect of education on fertility. These studies varysubstantially in their methodology. However, in all cases but five,both husband's and wife's education are included, and most use linearestimation of the relationship between education and fertility. Ofthese, 19 include husband's education. In six cases, the relationshipis negative and significant, whereas in two cases it is positive andsignificant; in the rest it is insignificant. The two perverse casesare for a group of urban men in Brazil in 1960 and for 717 heads ofhousehold in Sierra Leone in 1966-67.

For women's education, 13 of the 26 estimates are negative andsignificant, and two cases are positive and significant. The twoatypical cases are the rural Laguna survey, in which a matchinganalysis of the same data revealed a negative significant value, and arural Bangladesh sample of 265 women in 1968-69. In the latter case,the positive relationship would be reduced from +0.13 to +0.03 if theeffect of female education on child mortality and of child mortalityon fertility were considered in the ordinary least squares estimates;it would be reduced from +0.42 to +0.26 if the mortality effect wereadjusted for in the two-stage least squares estimates.1

These linear estimates can also be used for other analyses. For

- 31 -

example, Table 7 (Panel A) shows a very clear relationship betweensample size and significance: for small samples (under 300) only 13percent of the results are negative and significant; for samples 300to 1,000, 36 percent; and for samples over 1,000, 54 percent.

Another conclusion to be drawn relates to the effect ofcontrolling for age at marriage.1 3 Table 7 (Panel B) shows that, ifage at marriage is included, 45 percent of the cases are inverse formales

TABLE 7 Significance of Linear Estimates of the Effect of Education onFertility, by Methodology Used

Panel A. Results by Sample Size

Sample Size

Under 300 300-1,000 Over 1,000

StatisticalSignificance Male Female Male Female Male Female Total

Negative significant 0 2 0 5 6 9 22Not significant 5 5 6 2 5 7 30Positive significant 1 2 0 1 1 0 5

Percent negativesignificant 0 29 0 63 50 56 39

Panel B. Results by Inclusion of Age at Marriage in Regression

Age at Marriage Age at MarriageIncluded not Included

StatisticalSignificance Male Female Total Male Female Total

Negative significant 6 11 17 0 6 6Not significant 9 10 19 7 4 11Positive significant 1 1 2 1 1 2

Percent negativesignificant 38 50 45 0 55 32

- 32 -

and females; if age at marriage is not included, none of the male and55 percent of the female results are negative and significant. Thusinclusion of age at marriage has a different effect on the sign andsignificance for males and females.

To summarize, with the best methodology and the largest samplesizes, most studies show inverse results. However, even with the bestmethodology, the proportion of inverse results does not rise above 60percent. This confirms what was suggested earlier: in an importantnumber of cases, education is not monotonically, much less linearly,related to fertility.

In 11 studies, the authors explore the nonlinearity in therelationship between education and fertility. Anderson (1979) uses asemilog specification, four authors use a quadratic term, and sixauthors use dummy variables for various levels of education. Tobriefly summarize these results, Anderson's formulation showed thatthe negative impact of female but not male education increased aseducation rose. Studies using a quadratic specification find thatfertility first increases with education and then decreases. Thelevel of education with maximum fertility (obtained by solving for themaximum point on the quadratic curve) varies from 2.8 years in thePhilippines to 10 in Sierra Leone. In studies using dummy variablesfor the different levels of education, the patterns observed varyamong age groups and regions within countries. In general, the impacton fertility is more negative at higher education levels. In somecases, no level of female education has a significantly negativeimpact (women over 45 in Korea, landless households in India, urbanhouseholds in Kenya, rural households in Egypt); in one case,education has a significantly positive effect (male primary educationin rural Egypt). Part of this uneven pattern of significance may bedue to small sample sizes in particular age and education groups.

It is not easy to get an overall picture of theeducation-fertility relationship from these studies. Averaging thecoefficients is inappropriate because of the nonlinearity of therelationship. In addition, given the variety of methodologies used,most estimates are "biased' in one direction or another from someideal effect. To explore the true relationship further in amultivariate context would require some explicit causal modeling ofthe channels through which education acts. Mason and Palan (1980)have done this for Malaysia, and Smock (1980:87) for Kenya. Bothstudies show that, as more channels are included, the direct effect ofeducation diminishes to zero; however, neither goes through the finalstep of adding up the effect of education through various channels.

Several papers have carried out rather complete path analyses offertility. Loebner and Driver (1973) have done such a study with datafrom Central India. In this study, although neither husband's norwife's education was directly related to children ever born, bothaffect fertility indirectly. Husband's education operates throughoccupation (0.267), husband's income (0.275), wife's age at marriage(0.130), family structure (0.131), and absence of husband and wife(-0.12 and 0.08). Among these, only age at marriage affects fertility(0.052); therefore, husband's education has a small indirect positive

- 33 -

path of 0.130 x 0.052 = 0.0068. The path between wife's education andfertility is more tortuous: direct through age at marriage 0.191 x0.052 - 0.0099, and more indirect through marriage duration 0.191 x-0.272 x 1.032 x 0.721 = -0.0387. Thus the net effect of femaleeducation through these indirect paths is -0.0288.

Another study traces the effects of education on current fertility(rather than children ever born) for a sample of Colombian women (Chiand Harris, 1979). The analysis is done separately for women of zeroparity, parities one to three, and parities of four or more. Thedirect effects of education are not significant for those of zeroparity, -0.109 for those of parities one to three, and -0.099 forthose of higher parities; the indirect effects are +0.025, -0.001, and-0.062 for the three groups. Thus the total effects are +0.025,-0.110, and -0.161, respectively. The principle channels throughwhich education acts indirectly are spouses' discussion about familysize; contraceptive attitudes, knowledge, and use; and infantmortality experiences. This study illustrates the shift ineducation's effects as family size increases, as well as theimportance of indirect effects.

Other Considerations

Stability of Educational Differentials Over Time

It is important to determine the stability of educationaldifferentials in fertility across time. Unfortunately, in very fewcountries is this possible. Many censuses do not report fertilitydifferentials by education controlling for age; even more rarely arethere two such censuses for one country. Since most WFS data arestandardized by marital duration rather than age, the educationaldifferentials cannot be directly compared with census data.

Three countries do provide data of this kind: the Philippines,1968 and 1973; Jordan, 1972 and 1976; and Egypt, 1960 and 1976. Theresults of these comparisons are given in Table 8, showing thateducation has a more negative effect on fertility at the later pointin all three cases. Although this is a very small amount of evidence,it does seem to be confirmed by much more tenuous comparisons ofcensus (age) data with duration data from comparable WFS surveys.14

Comparisons of CELADE data with WFS data for urban areas (orspecifically for city or metropolitan areas where available) alsoconfirm this pattern if women 25-44 are compared with women of allmarital durations in the WFS. Any attempt at a finer breakdown (women35-44 and durations 10-19 and 20-29), yields confirmation for thelonger durations only.15

Although these data are suggestive, retabulation of WFS data byage would considerably increase our ability to :iake these ccmparisonsacross tirae. This would permit testing of the hypothesis thateducation's impact increases over time until fertility is selativelylow, at which point ths absolute but not the relative differentialsdecrease.

- 34 -

TABLE 8 Relationship Between Education and Fertility at Two Points in

Time: The Philippines, Egypt, and Jordan

Panel A. Philippines, Children Ever Born to Women 40-44 (ESCAP, 1978)

Percentage PercentageDifference Difference

CEB from CEB of CEB from CEB of

Education 1968 Uneducated 1973 Uneducated

No school 6.11 6.85

1-4 years 6.49 +6.0 7.04 +2.85-7 years 6.77 +10.8 6.75 -1.5High school 5.93 -2.9 6.09 -11.1College 4.27 -30.1 4.57 -33.0

Panel B. Egypt, Standardized Average Parity (World Bank, 1981)


Parity from Parity of Parity from Parity ofEducation 1960 Uneducated 1976 Uneducated

Illiterate 4.21 4.21

Read and write 4.53 +7.6 4.37 +4.2Elementary 4.26 +1.2 3.66 -13.1

Secondary 3.59 -14.7 2.80 -33.5

Panel C. Jordan, Children Ever Born Among Women Married over 20 Years

(Rizk, 1977; Jordan WFS, 1979)


CEB from CEB of CEB from CEB of

Education 1972 Uneducated 1976 Uneducated

Illiterate 8.79 9.00

Primary and 7.86 -10.6 7.23 -19.7

secondary

Secondary+ 4.70 -46.5 4.50 -50.0

- 35 -

Other Evidence on Urban-Rural Differences in Fertility

A number of country-specific studies specifically address orindirectly control for residence in the analysis of fertility. Inaddition to these studies and the WFS data used by Rodriguez andCleland, there have been a number of other fairly comprehensivestudies of urban and rural differences in fertility.

Kuznets (1974) reviewed data on urban-rural variations infertility for the late 1950s and early 1960s, and Findley and Orr(1978) reviewed data from the period around 1970. These two studiesuse different methodologies, though they do share two similarities:(1) they examine the fertility of all, not just married, women; and(2) they introduce no controls except for age. Although differencesin methodology prevent close comparison, these studies agree on somepoints that give considerable insight into urban-rural variations.

Findley and Orr found that the average total fertility rate was4.95 for urban areas and 6.35 for rural areas, yielding a differenceof 1.4 and a ratio of 1.28. This absolute difference falls betweenthe two estimates found using data from Rodriguez and Cleland,16 butthe ratio of fertility rates indicates much larger differences thanthose found by Kuznets. Kuznets found a ratio of 1.09 for developingcountries as a whole, this ratio varying from 1.03 for sub-SaharanAfrica to 1.12 for Asia and 1.15 for Latin America. At least some ofthe contrast with Findley and Orr is due to differences inmethodology. Kuznets used children under 5 per 1,000 women ofreproductive age as his measure of fertility. Using this measure,differences between urban and rural areas would result fromdifferences not only in fertility, but also in age structure and childmortality, although it is possible that these latter two factorscancel each other out somewhat. To the extent that Kuznets' resultsare accurate for the period around 1960, it is also quite possiblethat urban and rural differences in fertility widened over thedecade. Further support for a widening gap is found in the fact that,in Western Europe, residential differences widened during the periodof fertility decline and then began to narrow in the post-World War IIperiod (United Nations, 1973:97).

Additional findings of interest from these two studies concernregional differences. Findley and Orr found that the larger the city,the lower the level of fertility in Asia and Latin America; however,this relationship did not hold for Africa. Kuznets found greaterurban-rural differences in Latin America than elsewhere in thedeveloping world, whereas in Africa urban fertility was in some caseshigher than rural. This pattern can be checked against the Findleyand Orr data from the 1970s, showing that African differentials(1.29-1.39) are midway between those of Latin America (1.56-1.43) andAsia (1.20-1.12). This change in pattern may arise from differencesin measurement, actual changes in differentials, or the effect ofselecting particular countries for study. Unfortunately, the latterpoint can not be checked on a country-by-coun.ry basis since Kuznetsdid not re,rt country-specific data. These varying results may alsoarise not from regional factors per se, but from differences in the

- 36 -

degree or pattern of urbanization in different regions. Thus one mustinterpret these regional patterns cautiously.

Relatively little insight into the effect of urbanization onfertility is gained from the multiple regression studies reviewedabove (see the Appendix Table) since they were not designed for thatpurpose. Only 11 of the 26 studies include residence or a proxy intheir analysis.17 In these studies, significant negative effects ofurbanization are shown in five cases and positive significant effectsin one; the rest are insignificant. Given the relatively small numberof cases, it is impossible to derive patterns of significance andnonsignificance from these data.

On the other hand, three multivariate studies yield some insightinto the channels through which residence operates. A study of CostaRica by Michielutte et al. (1975) shows that adding other variables(education, income, age at first conception, and church attendance)significantly reduces the effect of residence. Of these variables,the addition of income has the greatest effect on the association,reducing the correlation coefficient from -0.32 to -0.24; no othervariable had an effect even a third this size. Ketkar's (1979) studyalso shows that fairly large urban-rural differentials in fertility(0.85) are reduced to statistical insignificance when income,education, infant mortality, size of extended family, religion, andoccupation are added to a regression. The Loebner and Driver (1973)path analysis of fertility in India examines residence as well aseducation and shows no direct effect of residence. Although residenceaffects husband's income, family structure, and education, its onlyindirect effects on fertility are those through education, which isitself an indirect effect. Residence's effect is thus much smallerthan education's. It is somewhat surprising that residence does nothave an effect more directly through age at marriage; however, forthis particular sample, no significant effect of this kind exists. Onthe other hand, this sample is perhaps atypical in that even thezero-order correlations between residence and children ever born arevery small (0.039).

There are other studies giving some fragmentary evidence on thechannels through which urbanization may operate. However, it seemsmore efficient to examine broader issues in addressing the effects ofchanging socioeconomic circumstances on fertility. The propositionsbelow summarize these issues.

PROPOSITIONS

Education and Fertility

1. The effects of education on fertility vary according to whethermale or female education is considered and whether all or only marriedwomen are included, and with the degree of urbanization.

2. A major effect of education on fertility is through age atmarriage; thus the observed negative effect of education on fertilityis smaller in samples including only ever-married women than in those

- 37 -

including both married and unmarried women.3. Monotonic inverse relationships between education and

fertility are much less likely to be observed in countries having lowlevels of urbanization, per capita income, and daily caloric intake.Of these three factors, urbanization appears to be the mostimportant. This may explain the stronger observed negativerelationship between education and fertility in Latin American than inAsia.

4. Female education is more often inversely related to fertilitythan is male education: the negative effect across the total range offemale education is about three times greater than that of maleeducation after adjusting for many other factors, including spouse'seducation, occupation, and residence. Male education at low levels ismore likely to be positively related to fertility than is femaleeducation; however, this may well reflect a positive income effect ofeducation since income is more closely related to male than to femaleeducation.

5. In multiple regression studies, the observed magnitude of therelationship between education and fertility is highly dependent onwhether or not the regression equation includes variables throughwhich education operates. Although several recent studies haveillustrated this by including these variables, the relative importanceof the different channels is still unclear and probably varies amongcountries.

6. The cross-section effect of education on fertility is morenegative at later than at earlier points in time, at least untilfertility levels become very low. This topic needs more research,which would be facilitated by a simple retabulation of WFS data inmore comparable form to those of earlier surveys and censuses. Inaddition to comparing differentials at various points in time,research should attempt to determine what aggregate factors, such asdevelopment or income growth, affect the changes in differentials overtime.

7. Reanalysis of the Value of Children data could provide insightinto the extent to which education's effect operates throughdifferential values.

Residence and Fertility

8. Differences in fertility between urban and rural areas are smallerthan those between the least and most educated; however, this mayresult from the dichotomous classification of residence and may thusbe an artifact.

9. The greater the degree of urbanization in a country, thegreater the urban-rural differences in fertility. This may arise inpart from measurement problems since, as urbanization increases, thoseclassified as urban generally come from larger cities.

10. The mechanisms through which residence operates are lessclear than those through which education operates. The former aremore difficult to identify because of the need to separate out factors

- 38 -

specific to location (place factors), as well as those specific toindividuals and themselves correlated with place factors. The WFSanalysis of such community variables as access to markets and averageeducation levels offers some hope of eventually disentangling thesechannels; reanalysis of Value of Children data could also provideinsight into the extent to which residence operates through theperceived costs and benefits of children.

NOTES

1. For example, if education affects age at marriage, then theinclusion of age at marriage (M) in a regression betweenage-controlled fertility (F) and education (E) will have thefollowing bias. Let F - a + bE + cM and M - d + eE. The effectof education on fertility cannot be measured by b alone, but mustbe measured by b + ec. If c is negative and e is positive, thenthe coefficient of b will be biased upward. This explains theplus in column 3 of Table 1.

2. Including age and marital duration is equivalent to including ageat marriage; this will cause a positive bias in the educationcoefficient.

3. It is interesting to note that in no case was husband's educationthe most important variable in reducing the negative effect ofwife's education, and that, for rural Malays, controlling forhusband's education had the most important impact on increasingthe negative effect of wife's education.

4. Studies using the total education of husband and wife introduceother problems--such as nonlinearities and the differentialeffects of husband's and wife's education.

5. Two solutions to the problem of variables determinedsimultaneously with fertility are two-stage least squares andreduced-form equations.

6. Other background variables, such as religion, caste, and parentalstatus and education, cause less well-defined problems because thedirection of bias depends on the effect of the background variableon education and fertility.

7. Therefore, in countries where the timing of fertility isundergoing significant change, these variables will be aninaccurrate reflection of completed fertility.

B. The value obtained for countries in which both estimates areavailable is 2.1; for all countries, the average difference is 1.9for married women.

9. Rather surprisingly, Findley and Orr also found that thepercentage of the population in small cities (under 20,000) wassignificantly related to larger urban-rural differentials.

10. Rodriguez and Cleland give no explanation for the asymmetricaltreatment of male and female education in this respect.

11. This description glosses over very detailed, careful discussionsof the measurement of the variables and the choice of modelspec ification.

- 39 -

12. If income or wealth is controlled, female education issignificantly inverse to fertility in 46 percent of the cases, andmale education in 29 percent. If there is no control for incomeor wealth, female education is inversely related 58 percent of thetime and male education 20 percent. If wealth, income, orhusband's education is controlled, wife's education is inverselyrelated in 49 percent of the cases.

13. If age and marital duration are both in the equation, this isequivalent to including age at marriage.

14. The approximate comparisons are as follows: for Kenya, women45-49 in 1962 and over 45 in 1977-78; for Indonesia, women 45-49in 1971 and marital duration over 20 in 1976; and for Malaysia,women 35-44 in 1966-67 and duration over 20 in 1974.

15. This is not surprising since in most cases, the negative impact ofeducation increases over the life cycle.

16. For all women, the average difference was 1.6 and for ever-marriedwomen 1.2.

17. Many control for residence by having samples restricted to urbanor rural areas.

BIBLIOGRAPHY

Afzal, M., Z. Khan, and N. A. Chaudhry (1976) Age at marriage,fertility and infant-child mortality in a Lahore suburb. Part I,Pakistan Development Review 15(l):90-109. Part II, PakistanDevelopment Review 15(2):195-210.

Aghajanian, A. (1979) The relationship of income and consumption ofmodern goods to fertility: A study of working class families inIran. Journal of Biosocial Science 11:219-226.

Anderson, K. H. (1978) The Effect of Child Mortality on the Quantityand Quality of Children in Guatemala. Unpublished Ph.Ddissertation. North Carolina State University, Raleigh, N.C.

Anderson, K. H. (1979) Determination of Fertility, Child Quality,and Child Survival in Guatemala. Economic Growth CenterDiscussion Paper No. 332. Yale University, New Haven, Conn.

Banskota, K., and R. E. Evenson (in press) Fertility, schooling andhome technology in rural Philippines households. PhilippineEconomic Journal.

Bulatao, R. A. (1975) The Value of Children: Philippines, Vol. 2.Honolulu: East-West Population Institute.

Buripakdi, C. (1977) The Value of Children: Thailand, Vol. 4.Honolulu: East-West Population Institute.

CELADE and CFSC (1972) Fertility and Family Planning in MetropolitanLatin American. Community and Family Study Center, University ofChicago.

Chaudhury, R. H. (1977) Education and fertility in Bangladesh.Bangladesh Development Studies 5:81-110.

- 40 -

Chernichovsky, D. (forthcoming) Fertility behavior in developingeconomies: An investment approach. In J. L. Simon and P.Lindert, eds., Research in Population Economics. Greenwich,Conn.: JAI Press.

Chernichovsky, D., and 0. A. Meesook (1980) Regional Aspects of

Family Planning and Fertility Behavior in Indonesia. World BankMimeo. World Bank, Washington, D.C.

Chi, P. S. K., and R. J. Harris (1979) An integrated model offertility: A multivariate analysis of fertility differentials inColombia. Canadian Studies in Population 6:111-125.

Cochrane, S. H. (1979) Fertility and Education: What do we ReallyKnow? Baltimore, Md.: The Johns Hopkins University Press.

Davidson, M. (1973) A comparative study of fertility in Mexico Cityand Caracas. Social Biology 20:460-472.

De Tray, D. (1979) The demand for children in a 'natural fertility"population. Pakistan Development Review 18:55-67.

Encarnacion, J., Jr. (1974) Fertility and labor force participation:Philippines 1968. Philippine Review of Business and Economics11:113-144.

ESCAP (Economic and Social Commission for Asia and the Pacific)(1978) Population of the Philippines. ESCAP Country MonographSeries No. 5. Bangkok: United Nations.

Evenson, R. E., B. M. Popkin, and E. King-Quizon (1979) Nutrition,Work, and Demographic Behavior in Rural Philippine Households: ASynopsis of Several Laguna Household Studies. Economic GrowthCenter Discussion Paper No. 308. Yale University.

Findley, S. E., and A. C. Orr (1978) Patterns of Urban-RuralFertility Differentials in Developing Countries: A SuggestedFramework. Washington, D.C.: Office of Urban Development,Development Support Bureau, Agency for International Development.

Goldberg, D. (1976) Residential location and fertility. In R. G.Ridker, ed., PoPulation and Development. Baltimore, Md.: TheJohns Hopkins University Press.

Hermalin, A. I., and W. M. Mason (1980) A strategy for thecomparative analysis of WFS data, with illustrative examples. InThe United Nations Programme for comparative Analysis of WorldFertility Survey Data. New York: United Nations Fund forPopulation Activities.

Hull, T. H., and V. J. Hull (1977) The relation of economic class and

fertility: An analysis of some Indonesian data. PopulationStudies 31:43-57.

Indonesia, Central Bureau of Statistics (1978a) Indonesia FertilitySurveY 1976. Principal Report, Vol. I. Jakarta: Central Bureauof Statistics.

Indonesia, Central Bureau of Statistics (1978b) Indonesia FertilitySurvey 1976. Principal Report: Statistical Tables, Vol. II.Jakarta: Central Bureau of Statistics.

Iutaka, S., E. W. Bock, and W. G. Varnes (1971) Factors affectingfertility of natives and migrants in urban Brazil. PopulationStudies 25:55-62.

- 41 -

Jordan, World Fertility Survey (1979) Jordan Fertility Survey, 1976Principal Report, Vol. I. Amman: Department of Statistics.

Kelley, A. C. (1980) Interactions of economic and demographichousehold behavior. Pp. 403-470 in R. A. Easterlin, ed.,Population and Economic Change in Developing Countries. Chicago:University of Chicago Press.

Kelley, A. C., A. M. Khalifa, and M. N. El-Khorazaty (1980)Demographic Change and Development in Rural Egypt. Paperpresented at the Institute of Statistical Studies Seminar on theDemographic Situation in Egypt, Cairo.

Kenya (1966) Kenya Population Census, 1962. African Population, Vol.III. Nairobi: Statistics Division, Ministry of Economic Planningand Development.

Ketkar, S. L. (1978) Female education and fertility: Some evidencefrom Sierra Leone. Journal of Developing Areas 13:23-33.

Ketkar, S. L. (1979) Determinants of fertility in a developingsociety: The case of Sierra Leone. Population Studies 33:479-488.

Khan, M. A. (1979) Relevance of human capital theory to fertilityresearch: Comparative findings for Bangladesh and Pakistan.Research in Human Capital and Development 1:3-43.

Khan, M. A., and I. Sirageldin (1979) Education, income, andfertility in Pakistan. Economic Development and Cultural Change27:519-547.

Kiranandana, T. (1977) An Economic Analysis of FertilitYDetermination Among Rural and Urban Thai Women. Institute ofPopulation Studies Paper No. 20. Bangkok: ChulalongkornUniversity.

Knowles, J. C., and R. Anker (1975) Economic Determinants ofDemographic Behavior in Kenya. Population Employment WorkingPaper No. 28. International Labour Organization, Geneva.

Kogut, E. L. (1974) The Economic Analysis of Fertility: A Study forBrazil. Population and Employment Working Paper No. 7.International Labour Organization, Geneva.

Kuznets, S. (1974) Rural-urban differences in fertility: Aninternational comparison. Proceedings of the AmericanPhilosopical Society 118:1-29.

Lee, B. S., R. E. Paddock, and B. F. Jones (1978) Development of anEconometric Fertility Model for Less Developed Countries: The NewHome Economics Approach to Female Status and Age at Marriage andFertility. First Draft. Interim Report prepared for the Officeof Population Policy Development Division, Department of State,Agency for International Development.

Loebner, H., and E. D. Driver (1973) Differential fertility incentral India: A path analysis. Demography 10:329-350.

Malaysia, World Fertility Survey (1977) Malaysian Fertility andFamily Survey--1974. First Country Report. Kuala Lumpur:Department of Statistics.

Mason, K. O., and V. T. Palan (1980) Female E,ducation, Fertility andFamily Planning Behavior in Peninsular Malaysia. Unpublishedmanuscript. University of Michigan, Ann Arbor, Mich.

- 42 -

Michielutte, R., C. A. Haney, C. M. Cochrane, and C. E. Vincent(1975) Residence and fertility in Costa Rica. Rural Sociology40:319-331.

Navera, E. (1978) Home Investment in Children in Rural Philippines.Paper presented at the International Center for Research on WomenWorkshop, 'Women in Poverty, What Do We Know?", Elkridge, Md.

Rizk, H. (1977) Trends in fertility and family planning in Jordan.Studies in Family Planning 8:91-99.

Rodriguez, G., and J. Cleland (1980) Socio-Economic Determinants ofMarital Fertility in Twenty Countries: A Multivariate Analysis.World Fertility Survey Conference, Substantive Findings SessionNo. 5, London.

Rosenzweig, M. R. (1978) The value of children's time, family sizeand non-household child activities in a developing country:Evidence from household data. Pp. 331-347 in J. L. Simon, ed.,Research in Population Economics, Vol. 1. Greenwich, Conn.: JAIPress.

Sarma, M. T. R. (1977) The Economic Value of Children in RuralIndia. Economic Growth Center Discussion Paper No. 272. YaleUniversity, New Haven, Conn.

Sathar, Z. A. (1979) Rural-urban fertility differentials: 1975.Pakistan Development Review 18:231-251.

Smock, A. C. (1980) The Relationship Between Education and Fertilityin Kenya. Nairobi: Central Bureau of Statistics, Ministry ofEconomic Planning and Development.

Snyder, D. W. (1974) Economic determinants of family size in WestAfrica. Demography 11:613-627.

Stycos, J. M. (1980) Education, Modernity, and Fertility in CostaRica. Unpublished manuscript. International Population Program,Department of Sociology and Center for International Studies,Cornell University, Ithica, N.Y.

Thailand, World Fertility Survey (1977) The Survey of Fertility inThailand: Country Report, Vol. II, Report No. 1. Institute ofPopulation Studies. Bangkok: Chulalongkorn University.

United Nations (1973) The Determinants and Consequences of PopulationTrends, Vol. I. Department of Economic and Social Affairs.Population Studies, No. 50. New York: United Nations.

UN/WHO Advisory Mission (1969) Pakistan: Report on the familyplanning program. Studies in Family Planning 40.

U.S. Agency for International Development (1978) Patterns ofUrban-Rural Fertility Differentials in Developing Countries: ASuggested Framework. Washington, D.C.: Office of UrbanDevelopment, Development Support Bureau, Agency for InternationalDevelopment.

World Bank (1980) World Development Report, 1980. Washington, D.C.:World Bank.

World Bank (1981) Some Issues in Population and Human ResourceDevelopment in Egypt. World Bank Mimeo. Population and HumanResources Division, Development Economics Department, World Bank,Washington, D.C.

- 43 -

Wu, T.-S. (1977) The Value of Children: Taiwan, Vol. 5. Honolulu:East-West Population Institute.

Zurayk, H. C. (1977) The effect of education of women andurbanization on actual and desired fertility and on fertilitycontrol in Lebanon. Population Bulletin of the United NationsEconomic Commision for Western Asia 13:32-41.

APPENDIX TABLE Multivariate Studies of Education and Children Ever Born in Developing Countries, by Region

Esti- Education's EffectSamples mation and Significance

Region Location, Tech- Education (t values in parentheses) Other Variables Includedand Study Year, Size nique Measure (elasticities in brackets) (sign if significant) Coments

LATIN AMERICA

Iutaka Brazil OLS Husband's .077 (2.46) Age +, age at marriage -, major bias due toet al., (1959-60) levels city size +, migratory inclusion of age at1971 1,280 urban status +, color -, social marriage and social

males status +, father-in-law's status.status +, father's status.

Rogut, Brazil OLS Years HE WE Age -, marital duration +, major bias due to1974 (1960 census) income (varying sign), marital duration,

2,083 NE -.088 (2.93) -.035 (0.87) type of marriage which affects the3,287 S -.065 (5.66) -.030 (1.84) consensual - (NE and S), coefficient of wife's

2,872 E -.059 (3.85) -.058 (2.68) rural +, suburban. education more thanhusband's.a

Davidon, Mexico Step- Wife'sa years -.096 (0.80) Age at marriage -, Age at marriage1973 Mexico City wise desired number of causes education to

(1963-64) children +. be underestimated,269 women as does desired 4>

35-39 size. Authorrecognizes former

Venezuela Step- Wife's years -.101 (0.76) Age at marriage -, effect in text.Caracas wise desired number of(1963-64) children +, wife's work192 women status, husband's35-39 occupation.

Hichielutte Costa Rica OLS Wife's years -.38 b Residence -, income, Bias probably intro-et al., (n.d.) church attendance -, duced by age at first1975 292 women age at first conception - contraceptive usel

35 49 also perhaps by(clinic and sample design.control)

Stycos, Costa Rica OLS Wife's years -. 1 6b Years married +, mass Since ma.s inform--

1980 (1978) information -, modern tion depends on edu-966 women psychology. cation, education is

over 35 undetestimated.Adjustment not possiblewith given data.

Sterilized H8 WzNo (550) -.05 -.05 Years sinei union +, con- No residence control.

Yes (267) .00 _m17b su er durablea -, siblings+, urban church attendance,work history, parent'seducation, current work.

Anderson, Guatemala OLS Years H8 Woalth, location, wife's Does not include

1979 (1974-75) (in -. 008 (1.00) -. 029 (3.22) age 4. imputed wage for

657 house- CEB) 1-.0211 1-.0541 husband.c

holds

EanT AMIA

Kirananamna, Iftailand OLS Levels as us Age ma, mrital ducation +, If husbnd' educa-

1977 (1969-70) ideal fmily eia. +, tion is deleted,

1,546 women -. 155 (4.05) -. 059 (1.31) contraceptive uase +, wife's Is negative476 rural -. 097 (0.95) .049 (0.43) proportion dying +, and highly signifi-

1,068 urban -. 162 (4.09) -. 073 (1.70) proportion male +, cant.(urban) .d

Le et nI., Korea OLS Age 35-39 Hu Ws mortality +, mortality Education undere-ti-

1978 (1971) 0 0.29(2. 17) 0.23(2.27) *quared -, surviving sons mated due to the ln-

2,334 women 7-9 -0.08(0.66) -0.52(3.58) -, wife urban -, modern clusion of mortality,

35-49 10-12 -0.22(1.79) -0.64(3.17) objects owned, family child enrollment. and12+ -0.26(1.47) -0.45(0.88) structure, children In employment.

agriculture, children In

Age 40-44 nonagriculture +,0 0.01(0.07) 0.39(2.51) regional nortality,

7-9 -0.20(1.10) 0.08(0.28) children enrolled,10-12 -0.17(0.77) -1.07(3.00) children under 6 +,12+ -0.18(0.56) -0.95(1.41) regional unmployment,

demand for feaale labor,

Age 45-49 husband's unemploymnt.0 -0.36(1.59) 0.20(0.84)7-9 -1 15(3.87) 0.05(0 10)

10-12 -0.31(0.79) -0.24(0.45)12+ -0.96(1.99) -0.85(0.69)

APPENDIX TABLU 1cntinued]

Esti- Education's EffectSample: mation and Significance

Region Location Tech- Education (t values in parentheses) Other Variables Includedand Study Year, Size nique Measure Jelasticities in brackets) (sign if significant) COaments

Mason and Malaysia OLS Years Urban Rural Age, marital duration. Biased due to marital

Palan, 1980 (1974) Race E E duration, but see6,147 Malay .095 .127 text for expanded

Chinese -. 1 1 7 b _. 0 9 6 b discussion.Other -. 1 8 0 b .034

E2 E2

Malay -. 0 1 5 b - 016b

Cherniichovsky Indonesia OLS Wife Java-Bali Outer Islands Age +, age at marriage -, Education's effectand Meesook, (1976) Elementary -. 076(1.79) .083(1.75) marital status -, probably biased pos-1980 60,000 1-2.01 11.91 religion, work in home e, itively by age at

households Jr. Hi. Voc. .023(0.16) .061(0.44) work outside +, modern marriage, but knowsJr. Hi. Gen. -. 133(1.38) -. 313(3.42) durables -, household contraception has aSr. Hi. Voc. -. 479(3.61) -. 663(4.70) expea.I1ure +, perverse sign, asSr. Hi. + -. 960(7.92) -. 888(6.72) expenditure squared -, does work outside the

knows contraception +, home. Thus overallrural - (Java-Bali). direction of bias is

unclear.

Encarnacion, Philippines OLS Wife's years WE - . 2 1 4 ( 3 . 0 5 )e Age at marriage -, Biased due to inclu- as1974 (1968 NDS) WE2 - -. 040(3.56) duration (nonlinear), sion of age at

3,629 women family income marriage.under 50 (nonlinear) +/-.

Rosenzweig, Philippines OLS Years HE .022(0.65) Age of husband +, age of Husband's education1978 (1968 NDS) WE - -. 091(3.51) wife +, child wage +, may be biased slight-

1,830 women regional infant mortality ly due to predicted35-49 +, religion, contraceptive wagel wife's educa-

knowledge, predicted tion may be biasedhusband's wage, farm. due to contraceptive

knowledge.

Banskota Philippines OLS Years HE - -. 049(1.43) Infant deaths + 1963 Compare results withand Evenson, Laguna WE - .172(3.31) husband's wage, predicted next study. This

in press (1977) wife's wage -, predicted specification has far320 rural child wage +, full income, more variables endo-households land, home technology -, genous with respect

year married -, father's to education sofarm background +, results are moremother's farm background, biased.mother's nonfarm.

Navera, 1978 Philippines OLS Years HE - -.131(1.58) Wife's age at marriage, Wife's education(in Evenson Laguna WE - -.182(2.07) duration of marriage +, underestimated dueet al., 1979) (1977) household income, wealth, to age at marriage

320 rural and duration.househqlds

SOUTH ASIA

Chernichovsky, India OLS Husband's yeare HE - .058(0.85) Log mother's age +, age Wife's educationforthcoming (1968-69) Wife's literacy WE - .341(0.952) at marriage -, agricul- biased positively by

213 rural tural income +, unskilled age at marriage;households income +, skilled income -, husband's education(one village) biased downward due to

effect of his educationon income.

Sarma, India OLS Wife Landed Landless Age of wife +/0, child Careful effort to1977 (1970-71) Some primary -.247(0.89) -.165(0.48) death rate +/+, value of exclude endogenous

1,111 landed 1-.00291 (-.0038] livestock, cultivated variables. However,households Above primary -.599(1.66) -.431(0.82) area +, intensive if education affects505 landless [-.00401 [-.00361 development program, landownership, some

Matriculated -1.44 (2.57) -.464(0.72) health center in village selectivity bias may[-.0039] [-.00261 +/+, school in village, exist.

Husband factory in village,Primary or above .95 (6.90) 1.19 (5.36) household electricity -/0,

1 .14011 1 .17261 distance to town 0/-,value of farm instruments

Chaudhury, Bangladesh OLS Years HE WE Duration of marriage, Husband's, wife's1977 Dacca 1-5 -1.03 (2.60) -0.675(2.05) age at marriage, education entered

(n.d.) 6-9 -0.64 (1.22) -0.953(3.32) labor force status, separately. Under-1,130 women Secondary -0.548(1.00) -1.25 (5.04) income, exposure to estimates wife's

BA -0.446(0.78) -1.50 (5.56) mass media.f education due toMA -0.39 (0.67) -1.83 (6.67) inclusion of age at

at marriage.

Khan, 1979 Bangladesh OLS Years HE - -.06 (1.00) (Dead children) +/+, See comments on(1968-69) WE - .13 (2.17) (monthly income) +/+, next study.265 rural (wife worked), land ownedwomen 35-49 -/0, income adequacy -/-,who want no age of wife, aware ofmore children family planning, nuclear

family, age of wife atmarriage 0/+, education ofchildren +/0 d

APPENDIX TABLE (continued)

Usti- Bducation's BffectSamples mation and SignLficance

Reglon Location, Tech- Education (t values In parenthesen) Other Variables includedand Study Years size nique Measure e*lasticities in bracketsl (sign If significant) Coents

Khan and Pakistan TSLW Husband'a years us - .04(0.10) (Ntu br of dead children) Wife's educationSirageldin, (1968-69) Wife's literacy WE - -. 48(0.68) -, (income), (wife underestimated due1979 269 women worked) -, urban +, nuclear to other variables.

35-49 who +, land owned -, house Also possible biaswant no more owned +, wife's age +, age due to selection ofchildren at marriage -, know of those who want no

family planning -, lnme more children.adequacy -, education of 58e text.children -. d

De Tray, Pakistan OLS Years HE - .38(2.00) Blectricity +, house Only child mortality1979 (1968-69) H82 - -. 09(2.10) type +, rural, wife born is likely to bias

861 aen WE - -. 33(3.30) urban, cucrent a*e +, results.35-49 mortality +.

Afsal at Pakistan OLS Levels HE - -. 037 (0.47) Mortality rate, age of Age and maritalal., 1976 Labore WE a -. 234 (3.68) mother, marital duration duration are implicLt

(1973) +. control for age at674 meun marrage and bias may

exLit.aD

Sathar, Pakistan OLS Years HB W Age +, age squared -, Wife' education I1979 (1975) Total -. 017(2.10) -. 058(3.85) age at marrlge -_ probably underesti-

4,949 women Urban -. 022(1.83) -. 052(2.83) mortality +, use of mated due to ag* at15-49 Rural -. 012(0.99) -. 070(2.07) conttaceptlon +. mariage, mortality, and

use of contraception.

MIDDLE EAST

Zurayk, Lebanon OL8 Levels HE - -. 08(1.82) Duration of carrlage +, Education's effect1977 (1976) WE - -. 06(2.73) number of child deaths probably underestl-

16 villages +, religion -. mated due to1,050 men nortality.15-44

Kelley Egypt OLS Wife wife's age +, age Age at firstet al., 1980 (n.d.) Some primary (19%) .05(0.71) squared -, age at marriage and child

3,812 rural Primary + ( 48) .02(0.15) marriage -, electricity deaths cause educa-households Husband +. child death +. tion coefficient to

Soe primary (23%) .16(2.48) personal asasts, real be biased.Comp. primary (6%) .15(1.35) assets +.Above primary (88) .17(1.52)

Snyder, Sierra Leone OLS Levels HE - .21(3.71) Wife's working age, child Probable bias upward1974 (1966-68) WE - -.19(2.17) survival -, age +. by child survival.

228 women HE - .19(1.56) Wife's wage it working,35-49 WE - -.19(1.40) labor participation,

survival of children -,age, age squared -,child education.

Ketkar, Sierra Leone OLS Years HE - .016(1.11) Wife's age +, number Careful use of1978 (1966-68) WE - -.011(0.61) of adult females +, reduced form.

1,999 women comunity infant1979 WE - .06 (1.03) mortality +, religion,

WE2 - -.006(1.17) residence, occupation,migration.

Knowles and Kenya OLS Years WE - .0256(0.697) Years married +, urban -, None of the inde-Anker, 1975 (1974) nonfarm employment +, pendent variables

1,073 women breastfeeding, household that are dependentincome, land owned +, on education areother wife, community significant.family planning.

Kelley Kenya OLS Husband Child death +, income +, Probable bias due toet al., 1980 (n.d.) Comp. primary -0.32(-0.67) age + child deaths, but

401 urban Secondary -2.82( 1.83) female education isnuclear Wife not significant in

households Comp. primary -0.29(0.70) child death equation.Secondary -3.88(2.72) I

aAge at marriage is highly dependent on education, but more dependent on wife's education. The coefficients of husband's and wife'seducation an age at marriage in each region axe shown below:

NE S E

Husband's .144 .214 .244Wife's .325 .322 .305

All values are highly significant, but adjustment of education coefficients is not possible given the specification.bt values not reported, but result is aignificant at 5 percent level.cThis is the third specification. With an imputed wage added, husband's education has a positive coefficient and is marginally significantif the imputed wage is a reservation wage. In another paper, the author also used an imputed wage for the wife. In that case, wife'seducation had a value of -0.956, but a t value of only 1.55 (Anderson, 1978:146).

dVariables in parentheses are endogenous.eThis is not the regression equation focused on by Encarnacion. His focus was on a more complex equation (p. 125) which uses a thresholdtreatment. That treatment confirms the above: at low levels of education, there is a positive effect of education; at high levels, there isa negative effect. The threshold value is 2.75 years.fSignificance not reported for any of these variables.

Effects of Education on Fertility and...

Documents

Transcript of Effects of Education on Fertility and...